Prompt. Learn. Transform: How AI Is Rewiring the Way Lawyers Work

0:00Robert Eder:Okay, so welcome everybody to today's Legally Speaking podcast live, proudly sponsored by Clio. So let's get into it. AI has exploded across the legal industry for many lawyers, but it still feels overwhelming, risky, or simply not for them. Today's guest has made it his mission to change that. I'm joined by the brilliant Robert Eder, who's an intellectual property lawyer, legal tech educator and one of Europe's leading voices when it comes to AI prompting for lawyers. Robert designs legal automation solutions, teaches lawyers around the world how to use AI safely, effectively and creatively and he's trained hundreds of lawyers on responsible AI, advanced prompting and practical workflows and he's now taking his international advanced prompting techniques for lawyers program global.

0:51Robert Eder:So a very warm welcome to the show, Great to have you with us. Thank you very much, Rob. It's a pleasure to be here. Thank you very much. It's an absolute pleasure to have you on the show today. And before we kind of get too into advance into the AI and prompting side of things, for those who may be less familiar, could you tell us a bit about your professional background and how did AI end up becoming a big part of your legal career? Yeah, sure. So originally, I'm from Hungary, where I started my career specializing in IP law, which came totally by accident. But I used to study IT in secondary school to a level where I learned programming languages.

1:28And during my junior lawyer ages, I was always the go-to tech person in the firm. You can imagine why. And three years ago, I moved to Cyprus to work as an IP lawyer for a retail broker company. And it was very funny because naturally, there was a lot of situation that required more efficient process optimization. and I spent learning and developing custom tech tools and then I became obsessed with it. And now when generative AI hit, I realized that it will make a legal tech revolution because until that moment I think most of the people just talked about it, but I didn't see any innovation and it's very funny because it was not targeting lawyers, but this is a tool to process natural language.

2:15And since lawyers are the architects of the language, I realized that we are actually the native speakers of this technology, not the IT guys. So that's when I decided to keep dive deep into prompting. And this is my story.

2:30Robert Eder:Yeah, and it's incredible what you have been doing. I think the advanced level that you're sharing, but also keeping it really easy as possible for lawyers to understand. So adoption levels go up is great. So you're touching on it there, but just going a bit deeper. How did you decide to start teaching AI and prompting for lawyers rather than staying in purely legal practice or legal tech? Yeah, that's a good question. Actually, I always had the intention to teach even before Gen AI hit. But all my friends who attended PhD told me out of it because they said it sucks. And I wanted to find something more practical, market-based approach always.

3:08And it also came with the fact that I saw a dangerous trend with AI adaption. So lawyers in general are not regarded as tech experts, we can be honest. And I saw that a software came that they didn't understand and that they may, and I saw that many of them are afraid of it. So I wanted to give the power back to lawyers. And that's where I realized that if I teach a lawyer to prompt or just how to use it, how to be more technically literate, then I will give them not just the fish but the fishing rods to catch the fish if you understand what I mean.

3:44Robert Eder:Absolutely getting themselves sufficient so they can go out there and do do it themselves I think is really important and you've trained hundreds and hundreds of lawyers and so I guess in your experience from all of those teachings what are the most common mistakes lawyers tend to make when using AI? Yeah, actually, I see the same patterns, and they rather come from the human trait in general. So they try to treat AI as a shortcut, and in addition to that, they use it with generic prompts, which means that they give very bad prompts, they want to solve all the tasks instead of them, and then they result in catastrophe.

4:30So they should understand that this tool is here to increase our efficiency and it should also be handled with hectic criticism. Just like working with a junior lawyer, it comes inherently that to a junior lawyer, we explain the context in a very detailed form. And then when they deliver their output, we review it with a criticized eye and then we will accept that work or refuse. And this is the same we have to handle AI. But for lawyers, for some reason, they look at it as a tool that will solve everything instead of them, while they are ready to charge the same amount to the clients that they used to.

5:09And they don't use it properly.

5:12Robert Eder:Yeah, and, you know, that's the key, isn't it, with all of these tools? It's really understanding how to get the best out of it. And you mentioned sort of, you know, lawyers are sort of the architects. It leads nicely on to why do you believe lawyers are actually better suited for AI prompting than tech professionals? Yeah, that's a good one. I always tell this to my students that what are the main characteristics of lawyers? We are precise. We can explain to the client everything with clarity. And our job is to anticipate the other party's interpretation of our words. And this is actually exactly what we do with AI when we are prompting, because we have to be precise with our prompts, be very clear with the context, and we have to anticipate how the AI interprets our words.

5:58And imagine that we spend our entire career anticipating how a counterparty will misinterpret a clause. And it should come natural that we can also put this knowledge to AI prompting. and we're just learning a few key rules to keep on track we can very easily use ai and and know how our prompt will result in a bad output or in a good output at least this is my opinion

6:27Robert Eder:yeah no absolutely and one of the key things from today's discussion i know you're both keen to get across is some practical steps for lawyers from your trainings and workings that you've been doing So if a lawyer wants to start using AI tomorrow, what should be the very first skill or prompt structure they should learn? Well, this should be an easy question, but there are so many. Okay, so I would suggest to use the RTF format, which is the approbation of role task format. Because just start asking naked questions and start by defining the role. like if you are specialized in IP law like me. Let's say that you are an attorney from a top-notch law firm specialized in EUIPO proceedings.

7:19Then there is a task. You just simply say that I need to draft an opposition and then you outline the structure of the opposition, how would you draft it in general. And then the format, which is literally the desired outcome of the output that the AI gives to you. and in the meantime try to give as much context as possible i always say that we are not even doing prompt engineering but the truth verse should be context engineering because this is actually what will give a competitive edge to those who know how to use a proper or not and if you follow this raw task format method then you are already ahead of 90 of the users

7:59Robert Eder:in my experience and it's to your point about not being generic and that with that format you can be more specific you can provide hopefully better context and as a result obviously the outputs will be far greater and okay let's talk about some real use cases now and just general practical impact because what is one mind-blowing example where ai has actually helped your legal work in a way that you didn't even expect yeah that's a good one so you know this tool uh was created to process natural language and i think it does its best when you simply ask it to process natural lecture language or interpret context.

8:34And one of these was a NIST classification of a trademark application. Because what happens when an IP lawyer receives such a task? They sit down with senior lawyers, and if they don't understand what the client product is about, then they will brainstorm for several hours until they can outline a proper NIST classification from the predetermined list. And once I just simply uploaded the client's product sheets to the AI and asked it to classify it in accordance with NIS classification. And in order to restrict and prevent hallucination, I downloaded the NIS classification PDF documents from YPOL's website and told AI to stay within this framework.

9:23And in like 20 seconds, it gave me a very good classification. And within another 20 seconds, I already had a client communication to update the client about it. So that was amazing. And one more that was really mind-blowing. I developed the tool. I actually started my newsletter in Hungarian and I developed the framework for subscription and I realized I needed a privacy policy. And after using AI to develop the entire software, I taught AI that use the GDPR and the Hangar and Data Protection Authorities guidelines and give me a privacy policy based on the features of the products and the data it collects.

10:05And it gave me such an accurate privacy policy, even taking into account the features and the functions of the software, what it does, how it processes data. And actually, that was really mind-blowing to me.

10:18Robert Eder:It's incredible, isn't it? And I think the thing I'm taking from that as well is the speed of which you're getting those outputs as well. but quality and that's the key with all of this to hopefully provide some education to our listeners and watchers around the world that you know with the right frameworks with the right education that these tools can really sharpen your axe you can be far more efficient and you can genuinely just be a far sharper smarter lawyer okay this is all sounding great but what tool would you recommend for more people at the beginning beginners when it comes to ai and ai adoption to not get overwhelmed and then lead to inaction?

10:51Yeah, there are a lot of biases, but I recommend just to don't overcomplicate it to GPT cloud or Gemini, you don't need an expensive enterprise level to learn the logic. Start with the general basic models. They are the smartest reasoning engines available right now. And the difference is people can can be very biased about differences. But basically, at the beginning level, it doesn't take a difference. For example, just a good example i always tell i think it comes down to the to the debate of owning an iphone or an android so i prefer i always prefer chat gpt now gemini is very good as well but let's stay with chat gpt i have another colleague who always swears on a cloud and uh funnily he has an android and i have an iphone and so i think this is really the level of preference yeah the

11:42Robert Eder:one thing's for sure that they're all pretty good and they all can be super powerful if leveraged effectively what's your take from i mean i like chat gbt i think it's you know familiarity as well i i think sometimes it's good to have a tool and continue to try and get better with that tool so i'm continuously trying to up my game when it comes to chat gbt i'm always open to testing and being curious about other ones but i think where i feel like i'm making most progress and and learning the most is around chat gbt and that lends nicely actually to what i was going to ask you next in terms of most lawyers treat chat gbt like google they ask a question and then they hope for an answer you teach more advanced concepts like using the xml tags and answering level so without giving away your whole course which i know is really really good how does adding structure to a prompt change the legal output specifically uh yeah that's a good one and when i was a beginner I didn't do any structuring and when I started to dive deeper into it I learned the principle of using XML tags which is basically a fancy word for indicating in the IT language where a section starts and when it ends and also there is the so-called markdown language which is basically using hashtags for for title section headings and these are all recommended techniques by the developers of AI tools so open AI recommends to use markdown language to structure your prompts with ChatGPT.

13:07And Anthropic, the developer of Cloud, in its guidelines to Cloud recommends XML tags. And this makes a lot of sense because AI may also get confused by walls of text just like humans. Imagine if you just give a messy 50 pages contract and say a vergenic prompt and also some other instruction, it may believe that your instruction is part of the contract and it gets lost with it. And it's very good. It increases the efficiency of your prompt significantly if you divide into context, then instructions, and then refer to external documents. And it will also make your job a lot easier when you need to edit your prompt and go back and find easily which part you need to amend.

13:55So it prevents AF from getting confused and ensures that your output is laser-focused, basically.

14:01Robert Eder:yeah yep very much so and i love having that sort of you know what we call forensic folks focused almost um you touched on it earlier in the conversation around hallucinations because there is still a huge fear for many folks around hallucinations and indeed data privacy so you teach responsible ai how can a lawyer trust the output okay so this is a two-fold question because it has an AI part and a human part. And let me start with the human part. So we often blame AI for errors, but it's what my experience is when I see the blamers, that it's often the user. They provide vague prompts and then they are surprised that the AI doesn't give the desired output and it hallucinates.

14:45So a precise structure prompt always drastically reduces the error rate. For example, when they say that please refer the German civil law of the BGB about liability clauses and they only ask this one. AI was trained on something. It has a static data set and by default it wouldn't search the internet for you. And it will look for the BGB in the training data and refer to something that might be outdated. But I always tell this on my courses. If you explicitly reference the URL of that BGB and ask AI to open this and then find what I need, then they will do the natural language processing, which they are the best at with.

15:29Or you can just simply copy paste the text as well. And then they have the context window and they will not try to reach something unknown, but they will work within the context window and they will just do natural language processing. And the other part is the verification. So you have to distinguish between engine and the encyclopedia. And basically it depends on the task. So if you use AI for natural language processing, like find the indemnity clause in the text, it will be incredibly reliable because it was built for that. It's just processing the data you gave in. But the risk comes in when you ask for external sources, external knowledge, like cite a specific case from a specific year.

16:14and this is where it hallucinates because it tries to fill in the gaps uh even if it doesn't know the exact answer it will always try to fill in the gap and so my rule is that trust the linguistic ability it will be flawless but verify the facts and uh this is why i said it's too flawed because you have to always keep the human in the loop yeah absolutely and again thanks for giving some

16:37Robert Eder:some good examples and context to that we talked before around sort of preferred platforms i guess now i want to ask about sort of general ai versus more specialized legal tools because the most common question i think you get is around should i pay for specialized tools like a harvey or a lexus nexus plus ai or just use the chat gbt clawed gemini etc so where do you draw the line between needing a specialized tool versus just using a smart prompt uh yeah i i get that so often so because you know people are always looking at their budgets and yeah it's a big difference to pay 20 per month or a lot more.

17:15But it all depends on what you need. Do you need logic or data? So if you need to draft an email, summarize an agreement, restructure a clause, or just draft something from scratch and you know how to prompt it well, then it's logic. And JGPT or any Gen.EI tool is perfect for that. But if you need to find a specific case law, for example, from 1994 from a specific jurisdiction, then it's data. and Chagipit doesn't have that reliability. So this is when you pay for the specialized tools. But still you have to be able to differentiate because if you know how to prompt well and you don't need that specialized data, then don't pay the premium prices for the logic that you can get for cheap.

18:00This is my opinion.

18:02Robert Eder:Yeah, no, and thank you for sharing that. And of course, folks will know that we partner with Clio and Clio have got a tremendous amount of data with their recent acquisition, obviously at Vlex and Vinson and that's an incredible tool for folks if you're looking for sort of you know specialized data for sure. You've said lawyers don't need 70 % of AI tools isn't that an exaggeration? Yeah the complete sentence is if you know how to prompt properly then you won't need 70 % of the AI tools and I know it sounds high but look at the architecture So the most major of the legal tools are what developers call wrappers.

18:39Under the hood, they are using the exact same models, JetGPT, Cloud or Gemini. And these you can access for$20 per month,$20. And I developed several software automations that uses AI under the hood. And I know that the real know-how is how you prompt AI under the hood. but it's a high premium price to pay for predetermined prompts. And this is a... So the value they sell is that they have the pre-written prompt from you and you don't have to know how to prompt. But now there is the top 30 % above the 70. Tools with properly crazed load databases or complex workflow integrations that are absolutely necessary.

19:23And I love those tools. This is like PO. And this is where it really comes to a competitive edge, where they don't just use AI under the hood and pre-prompt it, but they have a proprietary database that lawyers wouldn't have access to publicly or through digital AI models. and also they connect workflows. So they do something that would require several extra steps for a human and not just with ChatGPT, but with other tech tools and they combine it to one software. And then this is where it's, I think, worth it. So the rule of thumb is if a tool brings you link data or context, like Quio, buy it.

20:03If you just write text, then learn to prompt and save your money.

20:06Robert Eder:Yeah, no, great, great tip. Thanks for sharing that. And I guess, you know, going on to legal prompting training now, because there are thousands of YouTube videos out there on AI. So why does a lawyer need a specific prompting for lawyers course? Yeah, well, basically, because for some reason, what I experienced from my workshops for lawyers, despite they know very well how to solve a legal issue, how to outline a strategy for contentious proceedings or whatsoever, it doesn't come for them that naturally to outline a prompt strategy with AI. And general prompting tutorials teach you how to write poems or plan a travel, make a nice marketing strategy.

20:54They optimize for creativity, they introduce how to use these tools, and that's nice. But we don't want creativity, we want consistency and we want strategy to solve a legal case for us. So my course, I spent hundreds of hours developing my course and watching generic tutorials. And my course teaches how to restrict AI, how to stop it from hallucinating, how to make it speak legal English, not market English. And that's a completely different skill set. And this is why I think for lawyers, I can call it a one-stop-shop solution.

21:28Robert Eder:I love it. and yeah it's fantastic and i know you've put a lot of time energy and effort into this and i guess in an extension of that question what is the difference then between general prompting and legal prompting yeah that's that's a very good question and i think uh this is where it comes uh to having a legal training in the first hand because somebody for example i let's let's come back to ip law there's a marketing specialist they deal with brands, but they have no clue about trademark laws. So if they ask about what can I do if somebody infringes on my brand, they would get a generic answer, but they wouldn't be able to verify it.

22:09They wouldn't be able to even go with further ideas, with strategies. And the lawyer who had a legal training, they would be able to verify if the strategy outlined by the AI is correct. they would be able to draw up a strategy. For example, in my case, there is a dispute with another party. And how I prompt AI? I start with building up a strategy. I prompt it to add extra steps to examine circumstances. And then I get a very good output that I can immediately utilize in my real workflow. So what I would say that a legal prompting comes to the ability to be able to verify the legal background and the validity of the output and also to be able to outline a legal strategy for the case that came to the lawyer.

23:10And this is what I present a lot of time in my course. It's very hands-on. And I show a lot of this kind of legal prompting, which might be not for everybody who is not specialized on a certain type of legal field. But I teach the principles how anybody who watches this, they will be able to utilize in their home practice.

23:31Robert Eder:Yeah. And that's the key, isn't it? We want everyone to hopefully get the benefit of this new wave of technology that's coming through. You know, sticking with trading, as we mentioned throughout, You've trained hundreds of lawyers in Hungary and now taking this global with your obviously English course. What's been the biggest surprise or light bulb moment you've seen when a skeptical lawyer finally gets AI to work for them? Yeah, actually, it's very funny about just to amend your previous statement. So I live in Cyprus and I already started training in Cyprus. I've already been to Berlin to train lawyers there as well.

24:04And I'm in discussion with UK universities as well. and wherever I went it turned out that it's not tied to nationality. It always comes to the human characteristics and they are the most surprised about the speed how AI reduces the speed of everything and I have a good game in my workshops I always dedicate an hour to solve tasks together and I gave a fight and I have a game where I give lawyers, I separate them to two groups, the participants and I give one group the ability to use AI to solve the task and to the other I forbid to use AI. And I give them a two-pages-long agreement and I simulate a real use case.